Professional Documents
Culture Documents
Evaluating The Treatment Effect of Inflation Targeting
Evaluating The Treatment Effect of Inflation Targeting
Abstract
We evaluate the treatment effect of inflation targeting in seven industrial countries that adopted
this policy in the 1990s. To address the self-selection problem of policy adoption, we make use of a
variety of propensity score matching methods recently developed in the treatment effect literature.
Our results show that inflation targeting has no significant effects on either inflation or inflation
variability in these seven countries. Further evidence from long-term nominal interest rates and
income velocity of money also supports the window-dressing view of inflation targeting.
r 2007 Elsevier B.V. All rights reserved.
1. Introduction
Inflation targeting has become a popular framework for the conduct of monetary policy
since the early 1990s. Compared to other targeting regimes (e.g., monetary or exchange
rate targeting), inflation targeting features an explicit target for inflation and greater
$
The authors would like to thank Editor Robert G. King, an anonymous referee, and Kevin Grier for valuable
comments. Any remaining errors are ours.
Corresponding author. Tel.: +1 405 370 4052; fax: +1 561 297 2542.
E-mail address: haichunye@gmail.com (H. Ye).
0304-3932/$ - see front matter r 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.jmoneco.2007.06.017
ARTICLE IN PRESS
2522 S. Lin, H. Ye / Journal of Monetary Economics 54 (2007) 2521–2533
1
See Bernanke et al. (1999), Svensson (1997), and Mishkin (1999).
2
This argument is due to Anna Schwartz. See Romer (2006, p. 532).
3
See Ammer and Freeman (1995), Mishkin and Posen (1997), Groenveld (1998), and Kuttner and Posen (1999).
ARTICLE IN PRESS
S. Lin, H. Ye / Journal of Monetary Economics 54 (2007) 2521–2533 2523
Table 1
Starting year of inflation targeting in industrial countries
2. Data
The data set for this study includes 22 major industrial countries examined for the years
1985–1999.4 It contains 321 observations.5 Most of the data are drawn from the
International Monetary Fund’s World Economic Outlook and International Financial
Statistics.6
Seven countries—Australia, Canada, Finland, New Zealand, Spain, Sweden, and the
United Kingdom—adopted inflation targeting during our sample period.7 Following the
identification strategy employed by Ball and Sheridan (2003), we define each country’s
starting time of inflation targeting as the first year in which a specific target or target range
was in effect.8 Similarly, following Ball and Sheridan (2003), we define the starting time of
constant inflation targeting as the first year in which a country had an unchanging target or
target range. This is a more restrictive definition of inflation targeting, which we will use to
check the robustness of our results. To avoid confusion, throughout the article we will call
the first definition of inflation targeting non-constant inflation targeting. Table 1 lists the
seven targeting countries and their starting years. According to the definition of non-
constant (constant) inflation targeting, we identify 45 (41) targeting observations and 276
(280) non-targeting observations in our data set.
4
These countries are: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece,
Iceland, Ireland, Italy, Japan, New Zealand, Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, the
United Kingdom, and the United States.
5
There are nine missing values in year 1999 for some variables in our data set, which is why we do not have a
total of 330 observations.
6
We have also drawn data from OECD Main Economic Indicators (for M1 money stock), Ghosh et al. (2003)
(for five-year central bank governor turnover rate, CBTOR5), Ball and Sheridan (2003) (for the starting dates of
inflation targeting), and Reinhart and Rogoff (2004) (for exchange rate regimes).
7
Two of the targeters, Finland and Spain, adopted the euro in 1999. We still treat them as targeters in year 1999,
for targeting may still have some lagged effect in this short period. Changing them to non-targeters only makes
our results stronger. Switzerland adopted inflation targeting in 1999, but we still consider it a non-targeter in that
year, following Ball and Sheridan (2003).
8
Since Ball and Sheridan (2003) use quarterly data, they define the starting time of targeting as the first quarter
in which a specific target or target range was in effect and the target had been announced publicly at some earlier
time.
ARTICLE IN PRESS
2524 S. Lin, H. Ye / Journal of Monetary Economics 54 (2007) 2521–2533
3. Methodology
units on their propensity scores, which are the probabilities of policy adoption conditional
on X and can be estimated using simple probit or logit models. A further assumption
needed to apply propensity score matching is the common support assumption ( p(Xi)o1),
which requires the existence of some comparable control units for each treated unit.11
Using propensity score matching, the ATT now can be estimated as:
ATT ¼ E½Y i1 jDi ¼ 1; pðX i Þ E½Y i0 jDi ¼ 0; pðX i Þ. (3)
We consider a variety of commonly used propensity score matching methods. The first
method is nearest-neighbor matching with replacement, which matches each treated unit to
the n control units that have the closest propensity scores. We use two nearest-neighbor
matching estimators: n ¼ 1 and 3. The second method is radius matching, which matches a
treated unit to the control units with estimated propensity scores falling within radius r. We
use a wide radius (r ¼ 0.03), a medium radius (r ¼ 0.01), and a tight radius (r ¼ 0.005). The
third method is kernel matching, which matches a treated unit to all control units weighted in
proportion to the closeness between the treated unit and the control unit. The last method is
the regression-adjusted local linear matching developed by Heckman et al. (1998).
This section estimates the treatment effects of inflation targeting (non-constant or constant)
on the level of inflation (CPIG), defined as the annual growth rate of CPI, and inflation
variability (CPIGSTD), defined as the standard deviation of a three-year moving average of
inflation, in the seven targeting countries. Fig. 1 illustrates the average inflation and inflation
variability in targeting countries and non-targeting countries from 1985 to 1999. There is a
clear downward trend in inflation (variability) during this period in both country groups.
Therefore, a naive comparison of pre-targeting inflation (variability) and post-targeting
inflation (variability) in targeting countries can lead to the false conclusion that inflation
targeting matters. Indeed, by looking at the figures, one can reasonably suspect that the low
inflation (variability) might be caused by some common uncontrolled factors that affect both
targeting and non-targeting countries. In the rest of this section, we will use propensity score
matching methods to estimate the treatment effects on the treated.
We first estimate the propensity scores using a probit model. The dependent variable is
the targeting (either non-constant targeting (NCIT) or constant targeting (CIT)) dummy.
We consider two groups of control variables.12 The choice of the first group is based on the
literature that inflation targeting should be adopted only after some preconditions are
met.13 We choose the following five variables: the lagged inflation rate (CPIG_1), broad
money growth (BMG), government fiscal balance as a share of GDP (CGGDP), a
11
Estimating ATE requires a stronger common support condition (0op(Xi)o1), which requires the existence of
both comparable treated units for each control unit and comparable control units for each treated unit.
12
It is important to note that the goal of estimating the propensity score is not to find a best statistical model to
explain the probability of policy adoption. According to the conditional independence assumption, it is not a
problem to exclude variables that systematically affect the targeting probability but do not affect inflation
(variability) in the probit regressions. See Persson (2001) for detailed discussions.
13
See, for example, Truman (2003).
ARTICLE IN PRESS
2526 S. Lin, H. Ye / Journal of Monetary Economics 54 (2007) 2521–2533
0.10 0.035
0.09 0.030
0.08
0.025
0.07
0.06 0.020
0.05 0.015
0.04
0.010
0.03
0.02 0.005
0.01 0.000
1986 1988 1990 1992 1994 1996 1998 1986 1988 1990 1992 1994 1996 1998
Fig. 1. Average inflation and inflation variability in targeting countries and non-targeting countries. (a) Inflation,
(b) Inflation variability. Notes: Solid line indicates targeting countries. Dash line indicates non-targeting countries.
CBTOR5 as an inverse proxy of central bank independence, and real per capita GDP
growth rate (GDPPCG). We expect the first four variables to be negatively correlated with
the probability of adopting inflation targeting, and the last one to be positively correlated
with the probability. The second group of variables is used to control for the likelihood of
choosing exchange rate targeting as an alternative framework for the conduct of monetary
policy. We include a fixed exchange rate regime dummy (FIX) and trade to GDP ratio
(OPEN) as a measure of openness to trade in this group.14 Since exchange rate targeting is
more attractive to countries that have already adopted this policy and to countries that are
more open to trade, we expect to see negative coefficients on these variables.
The results are reported in Table 2. The two columns in Table 2 correspond to our two
targeting dummies NCIT and CIT. Most estimated coefficients have the expected signs.
We find that the lagged inflation rate, BMG, central bank governor turnover rate, and
fixed exchange rate regime dummy systematically affect a country’s targeting decision. The
estimated coefficients on these variables are all negative and significant, meaning that
countries with higher previous inflation, higher money growth, lower levels of central bank
independence, or fixed exchange rate regimes are less likely to adopt inflation targeting.
Other variables are not significant. The overall fit of these two regressions is reasonable
with pseudo-R-squares around 0.22.15
Before applying the matching methods, we want to make sure that our treated units
and control units share the same support, so that they are comparable. We sort all
the observations by their estimated propensity scores and then discard all the control
units whose estimated propensity scores are lower than the lowest score among the treated
14
We use the de facto exchange rate classification proposed by Reinhart and Rogoff (2004). Reinhart and
Rogoff classify exchange rate regimes into five broad categories: hard peg, soft peg, managed floating, freely
floating, and freely falling. We consider the first two categories as fixed regimes. A regime is classified as freely
falling if the annual inflation rate is higher than 40%. However, in our sample, no country experienced such high
inflation. Therefore, the omitted category contains only managed floating and freely floating regimes.
15
A pseudo R-square around 0.2 is comparable to an ordinary least-squares (OLS) adjusted R-square of 0.7. See
Louviere et al. (2000) for detailed discussions.
ARTICLE IN PRESS
S. Lin, H. Ye / Journal of Monetary Economics 54 (2007) 2521–2533 2527
Table 2
Probit estimates of propensity scores
Dependent variable
NCIT CIT
Notes: Constant terms are included but not reported. Robust standard errors are reported in parenthesis. *, **,
and *** indicate the significance level of 10%, 5%, and 1%, respectively.
units.16 Thus, 78 out of 276 control units are discarded when we use non-constant inflation
targeting, and 93 out of 280 control units are discarded when we use constant inflation
targeting.
The matching results based on these new samples are presented in Tables 3 and 4.17
Table 3 reports the estimated ATTs on the level of inflation, and Table 4 reports the
estimated ATTs on inflation variability. The upper panel of each table shows the ATTs of
non-constant inflation targeting, while the bottom panel shows those of constant inflation
targeting. The first two columns of each table show the results from one-to-one-nearest-
neighbor and three-nearest-neighbor matching. The next three columns report the results
from radius matching, with radii ranging from 0.5% to 3%. Local linear regression
matching and kernel matching results are shown in the last two columns of each table.
The results are strong and robust. The estimated ATTs in Table 3 are all found to be
quantitatively small and statistically insignificant. The average estimated ATT across
different matching methods and definitions of targeting is only about 0.17% in terms of
the annual inflation rate.18 The results on inflation variability are similar. The estimated
16
See Persson (2001). Keeping these observations does not change our results.
17
Matching estimates are obtained by using Stata command PSMATCH2 developed by Leuven and Sianesi
(2003).
18
For radius matching with medium radius and tight radius, a small number of treated units are discarded, for
no controls can be found within these radii. Other matching methods have employed all the treated units.
ARTICLE IN PRESS
2528 S. Lin, H. Ye / Journal of Monetary Economics 54 (2007) 2521–2533
Table 3
Matching estimates of treatment effect on the level of inflation
Notes: A 0.06 fixed bandwidth and a biweight kernel are used for kernel and local linear regression matching.
Bootstrapped standard errors for ATT are reported in parenthesis. They are based on 500 replications of the
data. *, **, and *** indicate the significance level of 10%, 5%, and 1%, respectively.
Table 4
Matching estimates of treatment effect on inflation variability
Notes: A 0.06 fixed bandwidth and a biweight kernel are used for kernel and local linear regression matching.
Bootstrapped standard errors for ATT are reported in parenthesis. They are based on 500 replications of the data.
*, **, and *** indicate the significance level of 10%, 5%, and 1%, respectively.
ATTs are all quantitatively small, and some of them are even positive. Almost all of them
are insignificant. The only exception is the estimated ATT of constant inflation targeting
on inflation variability using kernel matching. However, it is only marginally significant at
ARTICLE IN PRESS
S. Lin, H. Ye / Journal of Monetary Economics 54 (2007) 2521–2533 2529
the 10% level and quantitatively small. Overall, the evidence from matching suggests that
inflation targeting has no significant impact on either inflation or inflation variability.19
In this section, we employ the same propensity score matching methods to evaluate the
treatment effects of inflation targeting on long-term nominal interest rates and income
velocity of money.20 Long-term nominal interest rates are often used by policymakers as
an indicator of inflation expectations.21 Empirical studies in the literature have also shown
that real interest rates are fairly stable at long horizons, and that most of the movements of
long-term nominal interest rates are due to changes in expected inflation.22 If inflation
targeting can effectively lower the public’s expectations about inflation and inflation
variability, then both the level and variability of long-term nominal interest rates should
fall. The income velocity of money has also long been of interest to monetary
policymakers. The breakdown of monetarism in the 1980s was largely caused by volatile
velocity, which made targeting monetary aggregates an unreliable framework for
conducting monetary policy. Investigating the variability of velocity thus can help to
shed light on the window-dressing view. If inflation targeting is merely window dressing, it
should not affect velocity variability.
Fig. 2 illustrates the averages of long-term nominal government bond rates, bond rate
variability (defined as the standard deviation of a three-year moving average of bond
rates), and velocity variability (defined as the standard deviation of a three-year moving
average of the ratio of nominal GDP to M1) in targeting countries and non-targeting
countries from 1985 to 1999. There is a clear downward trend in long-term nominal
government bond rates in both country groups, while the downward trends in the two
variability series are less obvious. However, we can again observe that the movements in
each one of these series follow a very similar pattern in both country groups.
Formal empirical results from matching are presented in Tables 5–7.23 Tables 5 and 6
report estimated ATTs on the level and variability of long-term nominal government bond
rates. The results are striking: all estimated ATTs in Tables 5 and 6 are positive. There is
even some evidence, though it is very weak, that inflation targeting actually leads to higher
and more volatile long-term nominal interest rates. Unless one is willing to believe that
there have been dramatic increases in the level and variability of long-term real interest
rates in the targeting group, the above evidence implies that inflation targeting has not
significantly lowered the public’s expectations of inflation and inflation variability. Similar
results can be found when we apply matching methods to velocity variability. A majority
of the estimated ATTs are positive and no single ATT is statistically significant in Table 7.
There is no evidence that inflation targeting leads to more stable money demand in
targeting countries.
19
In fact, all the estimated ATTs on inflation and most of the estimated ATTs on inflation variability are
insignificant even at the 20% level.
20
We would like to thank an anonymous referee for this suggestion.
21
See Goodfriend (1993, 1998) and Goodfriend and King (2005).
22
See, for example, Barr and Campbell (1997).
23
These matching results are based on the same estimated propensity scores used in Section 4. Again, control
units whose estimated propensity scores are lower than the lowest score among the treated units are discarded to
ensure comparability.
ARTICLE IN PRESS
2530 S. Lin, H. Ye / Journal of Monetary Economics 54 (2007) 2521–2533
13 1.8
12 1.6
11
1.4
10
9 1.2
8 1.0
7
0.8
6
5 0.6
4 0.4
1986 1988 1990 1992 1994 1996 1998 1986 1988 1990 1992 1994 1996 1998
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
1986 1988 1990 1992 1994 1996 1998
Fig. 2. Average long-term nominal interest rates, interest rate variability and velocity variability in targeting
countries and non-targeting countries. (a) Long-term nominal interest rates. (b) Long-term nominal interest rate
variability. (c) Velocity variability. Notes: Solid line indicates targeting countries. Dash line indicates non-
targeting countries.
Table 5
Matching estimates of treatment effect on the level of long-term nominal interest rates
Panel A. Treatment effect of non-constant inflation targeting on the level of long-term nominal interest rates
ATT 0.3881 0.5846 0.7776** 0.6898 0.3175 0.7439 0.7392*
(0.6033) (0.4759) (0.3852) (0.4846) (0.5703) (0.8443) (0.3826)
No. of treated 45 45 45 44 38 45 45
No. of controls 36 80 197 169 105 197 197
No. of obs. used 81 125 242 213 143 242 242
Panel B. Treatment effect of constant inflation targeting on the level of long-term nominal interest rates
ATT 1.1461* 0.9098* 0.7554* 0.7985 0.8382 0.7713 0.6680*
(0.6070) (0.5043) (0.4171) (0.4989) (0.5940) (0.8045) (0.3908)
No. of treated 41 41 41 39 35 41 41
No. of controls 33 74 184 150 115 186 186
No. of obs. used 74 115 225 189 150 227 227
Notes: A 0.06 fixed bandwidth and a biweight kernel are used for kernel and local linear regression matching.
Bootstrapped standard errors for ATT are reported in parenthesis. They are based on 500 replications of the data.
*, **, and *** indicate the significance level of 10%, 5%, and 1%, respectively.
ARTICLE IN PRESS
S. Lin, H. Ye / Journal of Monetary Economics 54 (2007) 2521–2533 2531
Table 6
Matching estimates of treatment effect on long-term nominal interest rate variability
Panel A. Treatment effect of non-constant inflation targeting on long-term nominal interest rate variability
ATT 0.1537 0.1957 0.2070** 0.1850 0.2246 0.1960 0.1886*
(0.1395) (0.1158) (0.1002) (0.1330) (0.1581) (0.1617) (0.0975)
No. of treated 45 45 45 44 39 45 45
No. of controls 34 82 196 168 103 196 196
No. of obs. Used 79 127 241 212 142 241 241
Panel B. Treatment effect of constant inflation targeting on long-tem nominal interest rate variability
ATT 0.1019 0.1750 0.1699 0.1797 0.0426 0.1782 0.1674
(0.1500) (0.1334) (0.1167) (0.1363) (0.1624) (0.1665) (0.1098)
No. of treated 41 41 41 40 35 41 41
No. of controls 71 74 183 147 111 185 185
No. of obs. used 30 115 224 187 146 226 226
Notes: A 0.06 fixed bandwidth and a biweight kernel are used for kernel and local linear regression matching.
Bootstrapped standard errors for ATT are reported in parenthesis. They are based on 500 replications of the data.
*, **, and *** indicate the significance level of 10%, 5%, and 1%, respectively.
Table 7
Matching estimates of treatment effect on velocity variability
Notes: A 0.06 fixed bandwidth and a biweight kernel are used for kernel and local linear regression matching.
Bootstrapped standard errors for ATT are reported in parenthesis. They are based on 500 replications of the data.
*, **, and *** indicate the significance level of 10%, 5%, and 1%, respectively.
6. Conclusions
problem ignored in previous empirical studies. Using propensity score matching methods,
we find that the average treatment effects of inflation targeting on inflation and inflation
variability are quantitatively small and statistically insignificant in these seven countries.
The treatment effects on long-term nominal interest rates and income velocity of money
are also found to be insignificant.
Our results should be interpreted carefully. First, although no non-targeter in our
sample has publicly announced any inflation targets, some of them do have policies very
similar to those of the targeters. For example, the United States is often considered an
‘‘implicit targeter.’’ Therefore, it cannot be concluded from our results that efforts made by
central banks to reduce inflation are unimportant. Rather, our results imply that central
banks’ deeds matter more than their words. The additional explicit announcement of a
specific inflation target seems to have very little effect on the outcomes. Second, as many
emerging market economies and transition economies have adopted inflation targeting
recently, it would be interesting to investigate the treatment effects in these countries. Our
results may not necessarily apply to these countries, for their economic and social
structures are very different from those of industrial countries. Finally, as Ball and
Sheridan (2003) correctly point out, an insignificant treatment effect cannot be used as
evidence against inflation targeting as long as no harmful effect is found. Inflation
targeting may have other benefits not investigated in our study.
References
Ammer, J., Freeman, R.T., 1995. Inflation targeting in the 1990s: the experiences of New Zealand, Canada, and
the United Kingdom. Journal of Economics and Business 47 (2), 165–192.
Ball, L., Sheridan, N., 2003. Does inflation targeting matter? NBER Working Paper 9577.
Barr, D., Campbell, J., 1997. Inflation, real interest rates, and the bond market: a study of UK nominal and index-
linked government bond prices. Journal of Monetary Economics 39, 361–383.
Bernanke, B.S., Launach, T., Mishkin, F.S., Posen, A.S., 1999. Inflation Targeting: Lessons from the
International Experience. Princeton University Press, Princeton, NJ.
Dehejia, R.H., Wahba, S., 2002. Propensity score-matching methods for nonexperimental causal studies. Review
of Economics and Statistics 84 (1), 151–161.
Ghosh, A., Gulde, A., Wolf, H., 2003. Exchange Rate Regimes: Choice and Consequences. MIT Press,
Cambridge, MA.
Goodfriend, M., 1993. Interest rate policy and the inflation scare problem: 1979–1992. Federal Reserve Bank of
Richmond Economic Quarterly 79 (1), 1–23.
Goodfriend, M., 1998. Using the term structure of interest rates for monetary policy. Federal Reserve Bank of
Richmond Economic Quarterly 84 (3), 13–30.
Goodfriend, M., King, R.G., 2005. The incredible Volker disinflation. Journal of Monetary Economics 52 (5),
981–1015.
Groenveld, J.M., 1998. Inflation Patterns and Monetary Policy: Lessons for the European Central Bank. Edward
Elgar, Cheltenham.
Heckman, J., Ichimura, H., Todd, E., 1998. Matching as an econometric evaluation estimator. Review of
Economic Studies 65, 261–294.
Johnson, D., 2002. The effect of inflation targeting on the behavior of expected inflation: evidence from an 11
country panel. Journal of Monetary Economics 49, 1521–1538.
Kuttner, K.N., Posen, A.S., 1999. Does talk matter after all? Inflation targeting and central bank behavior.
Federal Reserve Bank of New York Staff Report 88.
Leuven, E., Sianesi B., 2003. PSMATCH2: stata module to perform full mahalanobis and propensity score
matching, common support graphing, and covariate imbalance testing. /http://ideas.repec.org/c/boc/bocode/
s432001.htmlS. Version 3.0.0.
Louviere, J.J., Hensher, D., Swait, D., 2000. Stated Choice Methods: Analysis and Application. Cambridge
University Press, Cambridge.
ARTICLE IN PRESS
S. Lin, H. Ye / Journal of Monetary Economics 54 (2007) 2521–2533 2533
Mishkin, F.S., 1999. International experiences with different monetary policy regimes. Journal of Monetary
Economics 43, 579–605.
Mishkin, F.S., Posen, A.S., 1997. Inflation targeting: lessons from four countries. Federal Reserve Bank of New
York Economic Policy Review, 9-110.
Persson, T., 2001. Currency union and trade: how large is the treatment effect? Economic Policy 33, 433–448.
Reinhart, C., Rogoff, K., 2004. The modern history of exchange rate arrangements: a reinterpretation. Quarterly
Journal of Economics 119, 1–48.
Romer, D., 2006. Advanced Macroeconomics. McGraw-Hill, Irwin, New York.
Rosenbaum, R., Rubin, B., 1983. The central role of the propensity score in observational studies for causal
effects. Biometrika 70, 41–55.
Svensson, L., 1997. Inflation forecast targeting: implementing and monitoring inflation targets. European
Economic Review 41, 1111–1146.
Truman, E., 2003. Inflation Targeting in the World Economy. Institute for International Economics, Washington,
DC.